Programming thread

  • Want to keep track of this thread?
    Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading.
    Create account
I’m finding that many people in the tech industry / software don’t work in these fields, and therefore don’t believe that these jobs exist. They have zero theory of mind about anything outside of modern SaaS business.

“Programming has always been and will always be a place where a bug can just be solved by an over the air patch!” is their mentality.
tell that to cnc machinists where a programming bug can turn thousands of dallars of material into scrap metal, and there's no roll back or undo once the part is damaged.
 
I'm trying to branch out and learn a new skill. Any recommendation?

Been thinking AWS/cloud eng or ai slop maker/prompt engineer
AWS is pretty tedious, but half the jobs now require either Azure, AWS or GCP. It is all over-engineered, IMO, as the vast majority of projects can run on an ancient n-tier architecture. AWS is popular because they do partnerships (like MS) where they will recommend you as a supplier. These greasy backroom deals have made my life miserable as a developer.
 
tell that to cnc machinists where a programming bug can turn thousands of dallars of material into scrap metal, and there's no roll back or undo once the part is damaged.
Storytime:

This was pre-AI. Our then-client, an online ad agency, wanted to expand to scamming small businesses, and for that they needed an interface where the customer mark would buy ads (their existing marks were large corporations and all had a manager attached).

They designed a workflow where a customer could buy (say) 1'000'000 clicks for $3 per click, get billed, then immediately (within a minute or so) change his campaign to 10'000 clicks for $300 per click, and weird syncing issues would have the bid go through as 1'000'000 clicks for $300 each (actually not quite that, the bids they were making were all per showing and their top secret high performance bidding engine was doing the conversion, but you get the idea). The real costs would eventually sync back and show the customer he was massively in the red, "your campaigns are paused, please top up your balance", but by that time the company would be massively in the red, too; a malicious actor could do this and vanish (losing his initial payment). Depending on how insane their margins were (they didn't tell us, I was just making an uneducated guess -- they did tell us they were "much higher than you can imagine" and online advertising was a massive scam), with the constraints they set (min/max buy, min/max cost per metric), it could take from $1000 to $100'000 to completely wreck them.

Inb4 "just validate input lol": I don't remember but there was a reason for why their system was designed that way, and I had to draw them a chart to persuade them this one weird trick was at all possible. The bug/feature was in their enterprise interface, too, it just never came up there because enterprise campaigns moved at the speed of email and didn't change their parameters this spectacularly in any case. In the end we had to force a 1-hour input lag AND hire an extra person to talk to them full-time.

Then their brownoid owner said "who the fuck are these people, why aren't we doing it in-house", cut us off in the middle of a month and ordered the accountant not to pay out the remainder. (My senior partner shook them down.) When I went to their office to finalize the papers, their poor HR lady was asking a girl (on the phone) how to sort a list in Python and offering her $18000/year. They were going to hire two such $18000/year programmers to replace us.
 
Last edited:
I'm trying to branch out and learn a new skill. Any recommendation?

Been thinking AWS/cloud eng or ai slop maker/prompt engineer
Every job listing I've applied to these last ~6 months has had AWS as a requirement. If you're interested in DevOps then every single listing wants you to know Terraform/OpenTofu, Ansible, CI/CD, AWS, Azure and some form of network engineering. Thankfully the certs for all of those are pretty easy and cheap, but it takes time if you haven't messed around with declarative systems before.
 
Every job listing I've applied to these last ~6 months has had AWS as a requirement. If you're interested in DevOps then every single listing wants you to know Terraform/OpenTofu, Ansible, CI/CD, AWS, Azure and some form of network engineering. Thankfully the certs for all of those are pretty easy and cheap, but it takes time if you haven't messed around with declarative systems before.
Don't care for bs certs; just what to learn and maybe I'll make a simple project
 
AWS is pretty tedious, but half the jobs now require either Azure, AWS or GCP. It is all over-engineered, IMO, as the vast majority of projects can run on an ancient n-tier architecture. AWS is popular because they do partnerships (like MS) where they will recommend you as a supplier. These greasy backroom deals have made my life miserable as a developer.

What's really fucked is when you have to work with these small companies who could easily run their shit on a $5 VPS but someone at some point convinced them that AWS was the way to go for their site that gets maybe 1000 hits a day (maybe they saw all the Super Bowl ads or whatever) and now everything is an order of magnitude more expensive and more complicated than it needs to be. So far I have yet to successfully convince one of these places to just let us save them money (and ourselves headaches) by moving them to DigitalOcean or Vultr or whatever.

Fuck AWS. All my homies hate AWS. You'd literally have to pay me to use that shit because there's no way I'd ever use it for anything I build myself. But yeah, I guess if you want to get a web job, you might as well learn it.

I've never used Azure or Google but I imagine they're pretty similar in terms of headache (and cost).

Speaking of overpaying for stupid shit, GitHub is sharply raising the price of its Copilot shit to more closely match the reality of its cost.
 
Another general question for the thread, what do you do and how rigorous is your workload? I feel like the AI stuff kind of fucked everything up now that every manager thinks you can 10x a feature by simply having Claude do it.
I've heard with my own ears "Don't just say 'this will take too long to implement', instead think about how you can use AI to do it faster!"
Made me want to suck-start a shotgun.
 
Unless there is actual evidence of what you claim (not what some people reckon). I will go with the simpler explanation.

Many scenes that are outside the norm lean to conspiracy, a lot of the time this is because of valid reasons, but I've seen it go off the rails plenty of times. People start repeating other people's stories and then it becomes a fact. When you look into it, you find out, there is very flimsy evidence or none at all. I am generally fed up with hearing such stories because that what starts off this loop.

Furthermore, I've seen quite a few places now require a real recording / face-to-face to prove you are a real person, which have nothing to do with this industry e.g. recently US companies have been asking Korean applicants to say something negative against Kim Jung Un to weed out North Korean spies.

So it doesn't sound far-fetched at all. I have to go through a bunch of background checks for most of the work I do anyway.

I'm readdressing this comment because it's worth mentioning.

In Mercor's former privacy policy it explicitly stated:
“User data may be utilized to improve the quality of Mercor's models. For example, a selection of interviews may be used in a dataset to train models for talent evaluation.”
It also said:
“Worker acknowledges that their data, including interview recordings, transcriptions, resumes, and other information… may be sent to external services for evaluation purposes.”
And their newer privacy/cookies page says they collect:
  • “Interview recordings, transcriptions, and responses to questions”
  • “Profile photos and images from interviews”
These AI annotation websites are harvesting user data, exactly as I stated before. It's not a conspiracy theory.
 
Another general question for the thread, what do you do and how rigorous is your workload? I feel like the AI stuff kind of fucked everything up now that every manager thinks you can 10x a feature by simply having Claude do it.
I've heard with my own ears "Don't just say 'this will take too long to implement', instead think about how you can use AI to do it faster!"
Made me want to suck-start a shotgun.
My workplace has done some minor experiments with AI assisted QA pipelines, and effectively written it off as not worth investing time and effort into.
 
Another general question for the thread, what do you do and how rigorous is your workload? I feel like the AI stuff kind of fucked everything up now that every manager thinks you can 10x a feature by simply having Claude do it.
I've heard with my own ears "Don't just say 'this will take too long to implement', instead think about how you can use AI to do it faster!"
Made me want to suck-start a shotgun.

Web dev with a small agency, currently, and basically in unofficial management over the other devs. Haven't seen a huge impact due to AI currently but I think it will come more in the form of not getting new clients than in our existing clients screaming at us. There's been lots of discussion about AI but fortunately nobody's forcing it on us yet. I've told the people under me that they can use it for research and stuff (evaluating options for various libraries, quickly finding an option which is buried in documentation) but that copy-pasting whole code is not allowed and so far they seem to be compliant - nobody's tried to do a 20,000-line PR after a day of work or some shit like that anyway.

I'm kinda worried about the future, but I don't know what I can do about it, so I'm just doing my work for now and hoping it all works out in the end.
 
Another general question for the thread, what do you do and how rigorous is your workload? I feel like the AI stuff kind of fucked everything up now that every manager thinks you can 10x a feature by simply having Claude do it.
I've heard with my own ears "Don't just say 'this will take too long to implement', instead think about how you can use AI to do it faster!"
Made me want to suck-start a shotgun.
It is about the same for me. The org I work at is quite slack, though. I use opencode (free models) and Claude basically the entire day now.

On any reasonably large codebase you will run out of Claude credits that they give you monthly on the "Pro" subscription. I use opencode (it is free) and I am working out how to run local coding LLMs.
Local coding LLMs are about the same speed as me.,

The vast majority of my day was never hardcore programming; I could hack for 10 hours straight on a Go codebase for 8 hours straight and super-optimise everything.
Most of the stuff I do is boring architectural stuff and keeping the three rubes I work with in line.

People both vastly overestimate and underestimate the abilities of AI. It is really good at doing refactors and fixing dumb repeated copy-and-paste crap over the codebase.
Also, I had it set-up code coverage reports, vulnerability scans, etc. and it fixed a bunch of dumb NPM issues. It is good for suggesting solutions.
 
Last edited:
What's really fucked is when you have to work with these small companies who could easily run their shit on a $5 VPS but someone at some point convinced them that AWS was the way to go for their site that gets maybe 1000 hits a day (maybe they saw all the Super Bowl ads or whatever) and now everything is an order of magnitude more expensive and more complicated than it needs to be. So far I have yet to successfully convince one of these places to just let us save them money (and ourselves headaches) by moving them to DigitalOcean or Vultr or whatever.

Fuck AWS. All my homies hate AWS. You'd literally have to pay me to use that shit because there's no way I'd ever use it for anything I build myself. But yeah, I guess if you want to get a web job, you might as well learn it.

I've never used Azure or Google but I imagine they're pretty similar in terms of headache (and cost).

Speaking of overpaying for stupid shit, GitHub is sharply raising the price of its Copilot shit to more closely match the reality of its cost.
If you have enough traffic to need AWS, you have enough money to not need AWS.
 
What's really fucked is when you have to work with these small companies who could easily run their shit on a $5 VPS but someone at some point convinced them that AWS was the way to go for their site that gets maybe 1000 hits a day (maybe they saw all the Super Bowl ads or whatever) and now everything is an order of magnitude more expensive and more complicated than it needs to be. So far I have yet to successfully convince one of these places to just let us save them money (and ourselves headaches) by moving them to DigitalOcean or Vultr or whatever.
My experience in industry is that stuff is either overengineered/underengineered or a combination of both. Also, a lot of companies / IT departments play it safe by going for an established player instead of using a basic open source stack. If AWS goes down, it goes down for everyone, and it isn't your fault. If your VPS or server goes down, it is your problem, and you get the blame.

There have been a huge number of times where I could have Flask/<some other MVC microframework> + Postgres running in a couple of Containers on a VPS or, fuck it, even a Raspberry PI and serving a site and it would have been fine.
Fuck AWS. All my homies hate AWS. You'd literally have to pay me to use that shit because there's no way I'd ever use it for anything I build myself. But yeah, I guess if you want to get a web job, you might as well learn it.
Every job requires it now. Luckily, I can tell Claude / OpenCode what to do 95% of the time and not have to bother. I am watching YouTube most of the day.
I've never used Azure or Google but I imagine they're pretty similar in terms of headache (and cost).
Azure is basically a copy of AWS. Dunno about GCP.
Speaking of overpaying for stupid shit, GitHub is sharply raising the price of its Copilot shit to more closely match the reality of its cost.
If I start contracting again, I am going to just drop £10k on the top-end GPU for AI and have it running coding models in my cupboard.
 
I've heard with my own ears "Don't just say 'this will take too long to implement', instead think about how you can use AI to do it faster!"
Made me want to suck-start a shotgun.
AI is kind of an amplifier like any tool. And like any tool it can 100% be misused and amplify crap. So in theory it can help you deliver stuff faster, but it might be covered by brown marks all over it if you don't take the right level of care.

My work is riding the robot-cock relatively hard at the moment. I even had to attend some training for developers from OpenAI recently. A large portion of it was essentially a sales pitch saying you should use it in all the parts of development, from planning, to designing, implementing, testing and monitoring. All that sounds like to me is "spend all your money and time using our tool." It actually sounded like hell to me, even down to their little neopet shit.

You've not really wanted to suck-start a shotgun until your manager hands you an AI generated "plan" that is almost useless, and defends it when you point out it's made stuff up and overcomplicated everything.
 
I wasn't forced to use it yet, but last week my boss reccommended to me, that I should attend a vibecoding training.
I probably will attend, I'm interested in what kinds of lunacy the cultists spew now. I doubt I'll find any use for it though.
I'm shocked that the term "vibecoding" gets unironically used, and not as an insult. Oh, well, clownworld and all that.
 
I finally broke down and asked the robot for help. Not because I needed help but I hate doing GUIs. I have a small 2" display and RP2350 chip I want to display stuff via HTTP POST. i figured this was a good test. Locally I have Qwen 3.6 27b and I have Claude Opus 4.7. The Qwen code is incomplete but a good starting point. The Clod Code is much more complete. Asking both for an addition seems to have worked. Clod gave me all the other files that need editing and not just the one I uploaded(I flushed context between requests) Qwen sort of hinted that other changes were needed. The Clod side cost me $0.91

I haven't yet tried the code on the device, we'll see what happens.
 
Last edited:
Back
Top Bottom